Guizhou Miao Embroidery Multi-Task Annotations

# Guizhou Miao Embroidery Multi-Task Annotations This repository contains the cleaned annotations and train/validation/test splits used for the multi-task embroidery analysis study. ## Image source The raw images are sourced from the public Kaggle dataset **“Guizhou Embroidery”** by Jinhui et al.: - URL: https://www.kaggle.com/datasets/jinhui1234/guizhou-embroidery - The original Kaggle license is listed as **Unknown**. Because the original image license does not permit redistribution, **the raw images are not included in this repository**. This package provides only the cleaned annotations and split files generated by the authors. ## Files - `labels.json` — Train/validation/test annotations. - `excluded.json` — List of low-resolution or mislabeled samples removed during cleaning. ## Annotation format `labels.json` contains three splits: `train`, `val`, and `test`. Each entry is keyed by image filename (without extension) and has the following fields: | Field | Meaning | Values | |---|---|---| | `auth` | Authenticity | `0` = fake / non-authentic, `1` = authentic | | `pattern` | Pattern category | `0` = bianxiu, `1` = duixiu, `2` = maweixiu, `3` = shushaxiu, `4` = other | | `defect` | Defect presence | `0` = no defect, `1` = with defect | ## Dataset splits | Split | Images | |---|---| | train | 34,330 | | val | 736 | | test | 736 | All images were resized and center-cropped to 224 × 224 pixels. ## How to obtain the raw images Download the original images from the Kaggle dataset linked above. The filenames in `labels.json` correspond to the cleaned image names used in the study. ## License The annotation files (`labels.json`, `excluded.json`) and this README are released under the **Creative Commons Attribution 4.0 International License (CC-BY-4.0)**. The original images remain under the license terms of the Kaggle dataset “Guizhou Embroidery”. ## Citation If you use these annotations, please cite the original study: > A three-branch dynamic multi-task network for authenticity discrimination, pattern recognition, and defect recognition of Guizhou Miao embroidery.